[
https://issues.apache.org/jira/browse/HIVE-19237?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16494691#comment-16494691
]
Zoltan Haindrich commented on HIVE-19237:
-----------------------------------------
I think the reliance on the operator id should be retired in that codepart; the
current solution which saves the operatorid as old works.
Trying to keep the old operator id at some many places will bump into problems
after some time, eventually someone will write "another" clone method - not to
mention other problematic scenarios, like: cloned twice...
Another option would be to mark the interesting operator and the look it up
when it's needed.
IMO keeping any information about an operator's "history" is not something the
operator should carry...
> Only use an operatorId once in a plan
> -------------------------------------
>
> Key: HIVE-19237
> URL: https://issues.apache.org/jira/browse/HIVE-19237
> Project: Hive
> Issue Type: Bug
> Reporter: Zoltan Haindrich
> Assignee: Zoltan Haindrich
> Priority: Major
> Attachments: HIVE-19237.01.patch, HIVE-19237.02.patch,
> HIVE-19237.03.patch, HIVE-19237.04.patch, HIVE-19237.05.patch,
> HIVE-19237.05.patch, HIVE-19237.06.patch
>
>
> Column stats autogather plan part is added from a plan compiled by the driver
> itself; however that driver starts to use operatorIds from 1 ; so it's
> possible that 2 SEL_1 operators end up in the same plan...
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
